Add agent control features: prompt enhancement, pause/resume, context…

[iGotYourBackMr]

Added Agent Control Features

This PR introduces three key quality-of-life improvements for users:

1. Prompt Enhancement Button ✨

   • Adds "Enhance" buttons next to task description, context input, and update context fields
   • Uses the currently selected LLM to improve prompts in-place
   • Maintains workflow continuity by avoiding context switching

2. Pause/Resume Functionality ⏯️

   • Adds pause/resume controls for agent execution
   • Implements safe state management through _global_agent_state
   • Maintains browser context during pauses

3. Context Update System 🧠

   • Allows mid-operation context updates via text input
   • Integrates with existing agent state management
   • Supports LLM-enhanced context updates using the same model

Implementation Details

• State Management: Extended AgentState class with pause/resume capabilities
• UI Integration: Added Gradio components with proper state synchronization
• Safety: Ensured browser context persistence during pauses
• Performance: Maintained existing execution patterns while adding new capabilities

Note:
While this may not be the fanciest implementation, it's been tested (with Gemini only but should work in general) to work reliably in these common scenarios:

• Interrupting long-running operations
• Updating context during complex tasks
• Enhancing prompts across different LLM providers

Notes for Maintainers

• Backwards compatible with existing workflows
• No additional dependencies required
• Includes error handling for state transitions
• Follows existing patterns in the codebase

Screenshot :

(Worth noting, I have used Cursor to make this work, I am no programmer myself but I am learning)

[CLAassistant]

All committers have signed the CLA.